Revision 0efca18 by Jeroen Bakker March 16, 2021, 14:03 (GMT) |
Cleanup: Modernize iter. |
Revision 5266e17 by Jeroen Bakker March 16, 2021, 14:03 (GMT) |
Cleanup: Modernize iterators. |
Revision 4ca8e09 by Jeroen Bakker March 16, 2021, 14:03 (GMT) |
Cleanup: Replace std::vector with blender::Vector. |
Revision 93e2491 by Jeroen Bakker March 16, 2021, 14:03 (GMT) |
Cleanup: replace std::map with blender::Map. |
Revision d07a769 by Sybren A. Stüvel March 16, 2021, 13:10 (GMT) |
Fix T86124: Self-hosting external libraries packages Allow downloading of source packages of Blender's dependencies, so that it's easier to provide a "full source archive" that contains the blender source + all dependencies archives. A `make` command for this will be introduced soon. This changes the deps builder slightly to be more flexible with the origin of our source packages. To support this a new CMake variable has been added called `PACKAGE_DIR` where all sources archives will be stored. default: a directory called `packages` in the build folder. alternative-default: if a directory called `packages` exists in the blender source folder that will be used. This is to support the "full source archive" use case. The download phase have been moved from the build phase to the configure phase. Configure will download all sources validate the hashes while downloading. All `[depname].cmake` files have been changed to take a local `file://[path_to_local_tarball]` path rather than a remote URI. A second requirement was that there needed to be an option to grab the sources from the blender SVN mirror rather than upstream. For this an option has been added PACKAGE_USE_UPSTREAM_SOURCES (default ON). The exact location in SVN still needs to be worked out, I tested with my local webserver and codewise it checks out. The path that is in there currently will not work (given there is no mirror there yet). To build this mirror our local package caches can be used. Reviewed By: lazydodo Differential Revision: https://developer.blender.org/D10598 |
March 16, 2021, 12:10 (GMT) |
Fix T86168: Add primitive Grid. Wrong number of subdivisions Changes to increase subdivision by one along both axis (X and Y) For example with x_segment = 3 and y_segment = 3. There should be 16 vertices ((3 + 1) * (3 + 1)) for correct number of subdivisions. Currently they are 3 * 3 = 9 vertices. Ref D10699 |
Revision cb3005c by Jeroen Bakker March 16, 2021, 09:12 (GMT) |
Fix: read cryptomatte manifests when images aren't loaded yet. A check was preventing the actual image to load and was therefore not parsed. |
Revision 6d011d0 by Jeroen Bakker March 16, 2021, 09:12 (GMT) |
Fix: Cryptomatte load corrupt manifests. Seems like one of the example files on the cryptomatte github is malformed and blender crashes when loading that file. This change will try to load as much as possible from the manifest so it can still be used. This has also been reported to cryptomatte project. |
Revision 0a34fec by Campbell Barton March 16, 2021, 08:30 (GMT) |
Fix T86561: Edit-mode crash with multiple objects sharing a mesh Use a for loop that always begins with the active object, instead of moving the active object in the array, which failed when it's data already being handled. While the existing logic could have been fixed, it's simpler to change the loop order. |
Revision d49e7b8 by Jeroen Bakker March 16, 2021, 06:43 (GMT) |
Compositor: Redesign Cryptomatte node for better usability In the current implementation, cryptomatte passes are connected to the node and elements are picked by using the eyedropper tool on a special pick channel. This design has two disadvantages - both connecting all passes individually and always having to switch to the picker channel are tedious. With the new design, the user selects the RenderLayer or Image from which the Cryptomatte layers are directly loaded (the type of pass is determined by an enum). This allows the node to automatically detect all relevant passes. Then, when using the eyedropper tool, the operator looks up the selected coordinates from the picked Image, Node backdrop or Clip and reads the picked object directly from the Renderlayer/Image, therefore allowing to pick in any context (e.g. by clicking on the Combined pass in the Image Viewer). The sampled color is looked up in the metadata and the actual name is stored in the cryptomatte node. This also allows to remove a hash by just removing the name from the matte id. Technically there is some loss of flexibility because the Cryptomatte pass inputs can no longer be connected to other nodes, but since any compositing done on them is likely to break the Cryptomatte system anyways, this isn't really a concern in practise. In the future, this would also allow to automatically translate values to names by looking up the value in the associated metadata of the input, or to get a better visualization of overlapping areas in the Pick output since we could blend colors now that the output doesn't have to contain the exact value. Idea + Original patch: Lucas Stockner Reviewed By: Brecht van Lommel Differential Revision: https://developer.blender.org/D3959 |
Revision 269536d by Campbell Barton March 16, 2021, 06:34 (GMT) |
Cleanup: use doxy sections for 'bpy.props' module |
Revision 95bb41a by Campbell Barton March 16, 2021, 06:22 (GMT) |
Cleanup: use _fn suffix in bpy.props |
Revision 81178ec by Campbell Barton March 16, 2021, 04:18 (GMT) |
Revert removal of lambda usage for Python RNA callbacks This reverts commits - 476be3746e85b4891189c8d480501905b9400c66 - 8d50a3e19e025ef470132e7edadd7b180db833f5 - 08dbc4f996e4e95f3ab64f7bb3e1193700c585f5 (partially). |
Revision e125305 by Campbell Barton March 16, 2021, 04:17 (GMT) |
Fix T86332: Error using lambda in annotations in Python 3.10 Callbacks used in `bpy.props` didn't hold a references to the functions they used. While this has been the case since early 2.5x it didn't cause any problems as long as the class held a reference. With Python 3.10 or when using `from __future__ import annotations`, the annotations are no longer owned by the class once evaluated. Resolve this by holding a reference in the module, which now supports traverse & clear callbacks so the objects are visible to Python's garbage collector. Also refactor storage of Python data, moving from an array into a struct. |
Revision be51d67 by Brecht Van Lommel March 15, 2021, 19:47 (GMT) |
Fix T86121: Cycles Attribute returning wrong results with OSL Fix uninitialized variable in the OSL shader. |
Revision 8f93386 by Brecht Van Lommel March 15, 2021, 19:46 (GMT) |
Fix (apparently harmless) Cycles asan warnings |
March 15, 2021, 17:01 (GMT) |
Cycles: optimize ensure_valid_reflection(), reduces render time by about 1% This is an implementation that is about 1.5-2.1 times faster. It gives a result that is on average 6� different from the old implementation. The difference is because normals (Ng, N, N') are not selected to be coplanar, but instead reflection R is lifted the least amount and the N' is computed as a bisector. Differential Revision: https://developer.blender.org/D10084 |
Revision fbe0165 by Mikhail Matrosov / Brecht Van Lommel March 15, 2021, 16:49 (GMT) |
Fix T56925: Cycles banding artifacts in dense volumes Offset the starting point of segments by a random amount to avoid the bounding box shape affecting the result and creating artifacts. Differential Revision: https://developer.blender.org/D10576 |
Revision cd3fade by Brecht Van Lommel March 15, 2021, 15:47 (GMT) |
Fix Cycles rendering crash on OpenBSD Static initialization order was not guaranteed to be correct for node base types. Now wrap all initialization in accessor functions to ensure the order is correct. Did not cause any known bug on Linux/macOS/Windows, but showed up on this platform. |
Revision 3fdef12 by Charlie Jolly March 15, 2021, 15:36 (GMT) |
Geometry Nodes: Support drag & drop object to create Object Info, Collection Info and Sample Texture nodes See: {T86296} Reviewed By: JacquesLucke Differential Revision: https://developer.blender.org/D10648 |
|
|
|


Master Commits
MiikaH:n Sivut a.k.a. MiikaHweb | 2003-2021